Académique Documents
Professionnel Documents
Culture Documents
In the United States and United Kingdom, exit polls play a key role in shaping
election night coverage. For anyone who has tuned into an Election Night television
broadcast in the U.S., much of the early coverage (and indeed much of the late) is spent
telling us "not only who won, but why they won". The "why they won" is a reference to
exit polls, which ask voters who they voted for and what shaped their vote. Often times,
the "why" question can provide valuable insight that would not be clear otherwise. In
2000, experts believed (and polls suggested) that Al Gore’s hopes of performing strongly
in Florida rested with seniors due to his "lock box" stand on social security, but the exit
polls showed that older voters split relatively evenly among the candidates. Instead, it
was young voters, happy about the strong state of the economy, who allowed Gore to end
up in a virtual dead-heat with George Bush in Florida. Few remember this age flip
revelation, but many recall the fact that the early exit polls indicated Al Gore would win
in Florida.
This half-right, half-wrong answer to whether the Florida exit poll did its job begs
the question: what is the fundamental purpose of an exit poll? Unlike in the U.S., U.K.
exit polling is used only to forecast results. Since 1997, British exit polls have accurately
estimated the winner of the British House of Commons election as soon as the polls
closed (see more below). Is it more important for an exit poll to explain a victory or to
predict one? What happens when exit polls disagree with the results of pre-election polls
or the actual results? In this two-part blog series, I intend to veer off my conventional
path and take a comparative look at past exit polls to try and answer these questions. In
1
this short blog entry, I will start off by looking at how British and U.S. exit polls differ. I
will show that British exit-pollsters conduct surveys with the chief goal of quickly
forecasting on election night which party (or no party) won parliament due to the fact that
election results are reported very slowly in Britain. My next post focuses on the
As the introduction suggests, exit polls in the United States serve the dual purpose
of early prediction, as well as explanation of the vote. These exit polls have been
audiences and providing valuable material to academics after the election. According to
the late head of Mitofsky International, Warren Mitofsky, raw horserace numbers, as in
Gore leads Bush 50-47%, are never purposely aired on Election Night (due to a relatively
high margin of error as well as the enthusiastic voters problem illustrated below). In the
United Kingdom, on other hand, exit polls have the sole purpose of predicting election
night outcomes. Like the United States, the exits polls in Britain are compiled for, and
reported by, media organizations. Unlike, in the United States, British media
organizations lay it out on the line and tell audiences that "the Conservatives are expect to
receive XXX seats in the House of Commons, while Labour is expected to receive
XXX". Also different from the United States, UK's exit polls do not aim at predicting
The reasons for these contrasts are rather clear. First, the British House of
2
course, polling 650 individual constituencies is essentially impossible, as it would require
well over 300,000 interviews to have a margin of error below +/- 5%. While is true that
presidential elections in the United States are of 50 statewide contests (less for senatorial
presents a major challenge. Due to the single member district rules of the British
parliament, a straight national vote to seat uniform swing cannot be utilized1. Thus, exit-
pollsters face the task of trying to make an aggregate prediction without being able to poll
every district. To get around this problem, exit pollsters (see Figure 1, page 6) interview
constituencies). That is, the first districts one would expect to flip to the opposition based
upon previous vote (gathered using exit polls as individual polling stations do not
typically report vote totals2) and demographic data. Using math far too complicated for
this post, the marginal constituency data is turned into a national seat estimate. Still, none
of this answers the question of why exit polls in Britain do not bother to find out "why"
The answer is both simple and complex. The pace of counting in each of the 650
different British constituencies differs considerably and often takes a very long time
because votes are counted by hand (vs. by machine in the United States). Even when
1 According to Mark Pack, a straight uniform swing had an average error among the three
main parties of 30 seats between 1997 and 2005.
2 See Curtice and Firth page 2 and Rallings et al. page 7. Most precinct level data is
available for only a limited time and is too resource intensive to gather quickly enough.
Some boroughs such as Brent, London keep precinct level data for longer. Otherwise,
exit poll and after election surveys such as the British Election Study are the only sources
for precinct level data.
3
Labour won 64% of the seats in Parliament in 1997, outgoing Prime Minister John Major
did not concede until 3:25 a.m., 5 1/2 hours after the polls closed. In the similarly unclose
2008 American presidential election, John McCain admitted defeat before midnight
Eastern time. Incoming Speaker of the House of Representatives, a large legislative body
like the House of Commons with 435 members, John Boehner also declared victory
before midnight in the lopsided 2010 election. Thus, the U.S media do not need exit polls
to declare winners at a relatively early hour3. In Britain, people watch the exit polls to
know the results before they go to sleep, according to exit pollster John Curtice. If exit
polls were not able to quickly project a winner, voters would not know who won the
The long exit poll questionnaires needed to determine why voters voted the way
they did would likely tamper the quick exit poll results. For example, exit pollsters in
Taiwan (see page 6044) found that shortening the questionnaire did lead to higher
response rates4. Those who would take the longer exit poll might be voters who were
more enthusiastic. In 2004 (see points 34 and 35), U.S. exit pollsters suffered from this
problem because Democratic voters were more eager to take the survey. The result was
skewed exit polls that over-predicted Democrat John Kerry's vote-share. Learning from
their early declarations in 2000, and understanding the possible survey errors associated
with exit polls (and any poll for that matter), the media did not report these pro-Kerry exit
polls. Instead, they waited for actual poll results to confirm the exit poll data. As we
know, it did not and the exit polls were then weighted accordingly to reflect the true
3Even in the tight 2004 election, NBC News was able to declare Bush a winner by 1 a.m.
4See Stephen Porter for a longer discussion of how shortening the survey length leads to
better response rates.
4
electorate.
The other issue in long questionnaires is one of time. In the United States, the
standard national exit poll asks over 50 questions. Reporting this large amount of data
from the field back to statisticians who can then weight and model it properly is a lengthy
process (see page 11). British exit pollsters are not afforded the time. According to exit
pollsters John Curtice and David Firth, Britons vote overwhelmingly in the late afternoon
and early evening. In fact, over 60% of British exit poll data are collected after 4 p.m.
Exit pollsters would not be able to transmit thousands of question laden late surveys to
properly reflect the late rush of British voters. Compare that to the United States, where
over 60% of exit data is collected before 3 p.m. American exit pollsters can take their
time collecting data over the course of a day, and if late day voters differ from earlier
ones, the media can simply say a race "is too close to call"5. In Britain, short surveys are
the only way to collect the data quickly and relay it quickly for top-line results to be
Concluding Thoughts
I must ask whether the rush for top-line results is really worth all the fuss? Is the
need for quick data so great as to sacrifice educational detail about the electorate that exit
polls like those in American supply? The obvious answer is to say that the media is free
to do what it pleases. One could also argue that more in-depth detail can be found in
election such as the British Election Survey. That said, many of these post election polls
5Exit poll data is reported at 12 P.M, 3 P.M, and just before the polls close in each state.
A composite of the first two “waves” is usually accurate enough to project winners in
non-tossup contests.
5
are not nearly as accurate as exit polls. Some voters have a tendency to lie about their
vote choice to fit the actual winner (in a type of bandwagon effect). Other people claim to
have voted when they had, in fact, not voted. Even if voters are honest, there is always
the possibility that a gap of time between casting a vote and answering a survey after
Election Day can impact a person's memory of why (s)he voted the way they did. Finally,
consider what would happen if these additional questions were added to the survey.
Fewer people would be questioned, and the margin of error for the top-line results would
increase. In the United States (unless the results are very clear or close), this causes the
media to wait 2-3 hours before announcing a result. If we apply that time frame to the
United Kingdom, an accurate seat count could still be predicted by 12 to 1 a.m. in the
morning. Perhaps that is too late, but that depends on what we want from the exit polls. If
the main reason for exit polls is to base a discussion around results, then the conversation
that could be generated by the results from the "why" questions asked in the exit poll
would also be entertaining and informative. Will we ever find out what type of exit poll